Page 1105
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
An Incident-Based Ethical Risk Screening Matrix for AI-Based
Information Systems Using Public AI Incident Records
Romelyn J. Banaybanay
1
, Reagan B. Ricafort
2
1
Initao College, Initao, Misamis Oriental, Philippines
2
AMA University, Makati City, Philippines
DOI:
https://doi.org/10.51583/IJLTEMAS.2026.150600080
Received: 21 June 2026; Accepted: 26 June 2026; Published: 07 July 2026
ABSTRACT
This study developed an incident-based ethical risk screening matrix for AI-based information systems using
100 publicly reported AI incident records from 2020 to 2026. It used descriptive content analysis and criterion-
based purposive sampling. Each incident was coded by year, system domain, ethical risk type, affected
stakeholder, ethical principle violated, severity level, and evidence strength. The study aimed to identify common
ethical risk patterns and translate them into a practical screening tool for IT managers and organizations. Results
showed that the most common ethical risks were misinformation or deception, followed by misuse or malicious
use, safety failure, privacy violation, and accountability failure. Generative AI and chatbots had the highest
number of incidents. Notable risk exposure was also found in law enforcement and surveillance, finance,
government and public service, education, and social media platforms. The most affected stakeholder group was
the general public. Most cases were assessed as high severity, while some were assessed as critical severity.
Another supporting pattern was the prevalence of deepfakes and synthetic media, mainly linked to
impersonation, fraud, misinformation, privacy harm, and reputational damage. The study concludes that AI-
based information systems need ethical screening early in adoption, deployment, or expansion. The proposed
matrix helps users identify warning signs, ask targeted screening questions, and apply management actions. It
also gives organizations a structured way to connect documented AI failures to practical review steps. The matrix
is not a substitute for a full technical, legal, or cybersecurity review. It uses documented AI failures as practical
evidence to support early risk identification and responsible IT management.
Keywords: AI ethics, AI incidents, ethical risk, information systems, screening matrix
INTRODUCTION
Artificial intelligence is now part of many information systems used in education, health care, finance,
employment, government service, transportation, social media, business, and public communication. These
systems process data, create content, support decisions, classify users, recommend actions, and automate
services. As use increases, ethical risks become easier to observe. These risks include misinformation, privacy
violation, bias, discrimination, safety failure, weak transparency, weak accountability, misuse, and harm to
vulnerable groups.
AI risk management has become a practical concern for organizations. The NIST AI Risk Management
Framework identifies safety, security, resilience, accountability, transparency, explainability, privacy, and
fairness as important characteristics of trustworthy AI [20]. The OECD also supports common reporting criteria
for AI incidents [22]. These frameworks show that AI risk is technical, ethical, and organizational.
Public AI incident records show how ethical risks appear in actual use. The AI Incident Database records real-
world harms and near-harms caused by deployed AI systems [1]. Prior studies use incident records to classify
risks, identify repeated failures, and improve AI incident reporting [7, 9, 16, 23]. Other studies use AI incident
cases to examine ethical issues, sector context, responsibility, and consequences [5, 11, 15, 30].
Page 1106
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
The AI ethics literature supports structured risk assessment. Researchers point to accountability, fairness,
transparency, human oversight, and harm reduction as basic requirements for responsible AI use [10, 24, 25].
Studies on ethics-based auditing show the need to turn broad principles into tools used in actual organizational
practice [18, 19].
Generative AI adds another layer of risk [3]. These systems create text, images, audio, video, and
recommendations. They also raise concerns about hallucination, misuse, user reliance, downstream use, and
synthetic media [4, 13, 29]. Deepfakes are included only as a supporting pattern because several encoded
incidents involved synthetic media. Studies on deepfakes show risks involving deception, identity misuse,
privacy harm, public trust, reputational damage, and disinformation [6, 12, 14, 17, 27, 28].
AI incident records and AI risk frameworks are available, but organizations still need simple tools that turn real
failures into early screening questions. Many frameworks explain risk at a broad level. Incident databases give
actual cases of harm. This study connects these two sources by using public AI incident records to develop an
ethical risk screening matrix for AI-based information systems.
This study analyzes 100 publicly reported AI incident records from 2020 to 2026. It uses descriptive content
analysis and criterion-based purposive sampling. Each incident is classified by system domain, ethical risk type,
affected stakeholder, ethical principle violated, severity level, and evidence strength. The results are then used
to develop the Incident-Based Ethical Risk Screening Matrix for AI-Based Information Systems.
The study is useful for IT managers, system developers, administrators, and organizations that plan to adopt or
use AI-based information systems. The matrix guides them in identifying warning signs, asking ethical screening
questions, and planning preventive actions before AI systems create harm.
Objectives of the study
The general objective of this study is to develop an Incident-Based Ethical Risk Screening Matrix for AI-Based
Information Systems using publicly reported AI incident records.
Specifically, this study aims to:
1. Identify the common ethical risk types found in publicly reported AI incident records.
2. Classify AI incidents by system domain, affected stakeholder, ethical principle violated, and severity level.
3. Analyze patterns among ethical risk types, system domains, affected stakeholders, and severity levels.
4. Develop an ethical risk screening matrix based on the classified AI incident records.
5. Recommend how IT managers and organizations should use the matrix for early ethical risk assessment.
Research questions
This study seeks to answer the following questions:
1. What ethical risk types appear in publicly reported AI incident records?
2. What AI-based information system domains are most commonly involved in reported AI incidents?
3. Who are the affected stakeholders in the reported AI incidents?
4. How severe are the ethical risks found in the incident records?
5. What ethical risk screening matrix can be developed from the classified AI incident records?
Page 1107
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
Scope and limitations
This study is limited to 100 publicly reported AI incident records from 2020 to 2026. It uses public records as
documentary data. It does not include system development, technical testing of AI models, surveys, or
interviews.
The study focuses on ethical risk classification and screening matrix development. It does not represent all AI
incidents worldwide. Because the sample was selected through criterion-based purposive sampling, the findings
are intended for descriptive analysis and practical risk screening.
The study depends on the completeness and accuracy of available public reports. Some records contain limited
technical detail. For this reason, the coding and severity ratings are based only on information found in the public
records.
METHODOLOGY
Research design
This study used descriptive content analysis. It examined publicly reported AI incident records to identify
recurring ethical risks in AI-based information systems. This design was suitable because the study organized,
interpreted, and classified documented AI incidents by ethical risk pattern, stakeholder, system domain, and
severity level.
Descriptive analysis summarized the frequencies and patterns in the dataset. These results became the basis for
the Incident-Based Ethical Risk Screening Matrix.
Data source
The study used publicly reported AI incident records as its primary data. The main source was the AI Incident
Database, which records real-world harms and near-harms caused by deployed AI systems [1]. The OECD AI
incident reporting framework, MIT AI Risk Repository, NIST AI Risk Management Framework, and AI incident
reporting studies supported the coding structure [7, 20, 22, 26].
The dataset consisted of 100 AI incident records from 2020 to 2026. The records came from AIID incident IDs
1254 to 1353. These IDs were part of a documented AIID batch of 108 new incidents reported in the November
2025 to January 2026 roundup [2].
Sampling technique
The study used criterion-based purposive sampling because the analysis required incident records with enough
detail for ethical risk classification. The goal was to select relevant and usable cases, not to create a statistical
sample of all AI incidents.
The sample was limited to 100 AI incident records. This size was enough for frequency analysis, cross-
tabulation, severity assessment, and matrix development, while keeping the work manageable.
Inclusion and exclusion criteria
An incident was included if it involved an AI-based information system, was publicly reported, occurred or was
reported from 2020 to 2026, had enough detail for coding, and showed at least one ethical risk. These risks
included bias, privacy violation, misinformation, safety failure, accountability failure, lack of transparency,
misuse, security risk, overreliance, and social harm.
Page 1108
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
An incident was excluded if it did not clearly involve AI, lacked enough detail for classification, duplicated
another included incident, fell outside the selected coverage period, or described general AI concerns without a
specific reported case.
Unit of analysis and coding framework
The unit of analysis was one publicly reported AI incident. Each incident formed one row in the coding sheet.
Each incident was coded by incident number, year, incident title, AI system or platform, short description, source
link, system domain, ethical risk type, affected stakeholder, ethical principle violated, severity level, reason for
severity rating, and evidence strength.
The system domain categories were education, health care, finance, employment and hiring, law enforcement
and surveillance, government and public service, social media and online platforms, transportation, business and
customer service, generative AI and chatbots, and others. Ethical risk categories were bias and discrimination,
privacy violation, misinformation or deception, safety failure, accountability failure, lack of transparency or
explainability, misuse or malicious use, security risk, overreliance or harmful human-AI interaction, and social,
economic, or environmental harm.
The ethical principles used in the study were fairness, privacy, transparency, accountability, safety, human
oversight, security, and social responsibility. These categories were guided by NIST trustworthy AI
characteristics, AIID taxonomies, OECD incident reporting criteria, and MIT AI risk domains [1, 20, 22, 26].
Severity classification
The study used a four-level severity scale. Level 1, Low, referred to minor inconvenience, limited harm, or small
scope. Level 2, Moderate, referred to clear but limited or reversible harm. Level 3, High, referred to serious
harm, legal risk, financial loss, privacy exposure, discrimination, safety concern, or broad effect. Level 4,
Critical, referred to severe harm, major rights violation, widespread public impact, serious safety risk, harm to
vulnerable groups, or major loss of public trust.
Data collection and analysis procedure
The researcher identified eligible AI incident records from public AI incident repositories and related public
reports. Each selected incident was entered into the coding sheet. Duplicate records, unclear cases, and records
with insufficient detail were removed.
The encoded dataset was analyzed using frequency counts and percentages. Cross-tabulation was used to
examine patterns between ethical risk type and severity level, system domain and ethical risk type, affected
stakeholder and severity level, and ethical principle and risk type.
The frequency tables and cross-tabulations were used to develop the Ethical Risk Screening Matrix. The matrix
turned observed patterns into warning indicators, screening questions, risk levels, and suggested management
actions.
Ethical considerations
This study used publicly available records only. No private personal data were collected. The analysis focused
on incident descriptions, ethical risk categories, system domains, stakeholder groups, and severity levels based
on public documentation.
Source links were recorded for traceability. The study avoided unsupported claims and did not assign blame
beyond the public record. Severity ratings were based only on available evidence.
Page 1109
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
RESULTS
This section presents the results from the 100 encoded AI incident records. The analysis focused on year, system
domain, ethical risk type, affected stakeholder, ethical principle, severity level, and selected cross-tabulation
patterns. Table 1 presents the summary frequency distribution.
Table 1. Summary Frequency Distribution of the 100 AI Incident Records
Variable
Category
Frequency
Percent
Year
2025
80
80.0%
Year
2026
9
9.0%
Year
2024
6
6.0%
Year
2023
4
4.0%
Year
2020
1
1.0%
System Domain
GEN - Generative AI and chatbots
28
28.0%
System Domain
LAW - Law enforcement and surveillance
16
16.0%
System Domain
FIN - Finance
13
13.0%
System Domain
GOV - Government and public service
13
13.0%
System Domain
EDU - Education
10
10.0%
System Domain
SOC - Social media and online platforms
8
8.0%
System Domain
TRN - Transportation
5
5.0%
System Domain
BUS - Business and customer service
4
4.0%
System Domain
HLT - Health care
3
3.0%
Ethical Risk Type
MISINFO - Misinformation or deception
33
33.0%
Ethical Risk Type
MISUSE - Misuse or malicious use
26
26.0%
Ethical Risk Type
SAFETY - Safety failure
12
12.0%
Ethical Risk Type
PRIV - Privacy violation
12
12.0%
Ethical Risk Type
ACC - Accountability failure
11
11.0%
Ethical Risk Type
BIAS - Bias and discrimination
4
4.0%
Page 1110
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
Ethical Risk Type
SEC - Security risk
2
2.0%
Affected Stakeholder
PUBLIC - General public
60
60.0%
Affected Stakeholder
CUSTOMER - Customer or client
14
14.0%
Affected Stakeholder
CHILD - Children or minors
10
10.0%
Affected Stakeholder
ORG - Organization or company
4
4.0%
Affected Stakeholder
PATIENT - Patient
4
4.0%
Affected Stakeholder
VUL - Vulnerable group
3
3.0%
Affected Stakeholder
STUDENT - Student or learner
2
2.0%
Affected Stakeholder
USER - Individual user
2
2.0%
Affected Stakeholder
EMPLOYEE - Employee or job applicant
1
1.0%
Ethical Principle Violated
SECURE - Security
28
28.0%
Ethical Principle Violated
TRANS - Transparency
22
22.0%
Ethical Principle Violated
SAFE - Safety
12
12.0%
Ethical Principle Violated
PRIVACY - Privacy
12
12.0%
Ethical Principle Violated
ACC - Accountability
11
11.0%
Ethical Principle Violated
RESP - Social responsibility
11
11.0%
Ethical Principle Violated
FAIR - Fairness
4
4.0%
Severity Level
3 - High
75
75.0%
Severity Level
4 - Critical
21
21.0%
Severity Level
2 - Moderate
4
4.0%
Evidence Strength
MODERATE
89
89.0%
Evidence Strength
STRONG
11
11.0%
Year distribution
The dataset covered incidents from 2020 to 2026. Most records were reported in 2025, with 80 cases or 80%.
This was followed by 2026 with 9 cases, 2024 with 6 cases, 2023 with 4 cases, and 2020 with 1 case. No encoded
Page 1111
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
record in the final dataset was listed for 2021 or 2022. Since 2026 was still a partial year, its count should be
treated as preliminary.
System domain frequency
Generative AI and chatbots recorded the highest number of cases with 28 incidents. Law enforcement and
surveillance followed with 16 cases.
Finance and government/public service had 13 cases each. Education had 10 cases, social media and online
platforms had 8 cases, transportation had 5 cases, business and customer service had 4 cases, and health care had
3 cases.
These results show higher risk exposure in public-facing systems, surveillance systems, financial systems, and
public service systems. These domains involve public trust, personal data, safety, rights, and organizational
accountability.
Ethical risk type frequency
Misinformation or deception was the most frequent risk type, with 33 cases. Misuse or malicious use followed
with 26 cases. Safety failure and privacy violation had 12 cases each. Accountability failure had 11 cases. Bias
and discrimination had 4 cases, while security risk had 2 cases.
The high count for misinformation and deception is linked to generative AI, synthetic media, deepfakes, and
automated content generation. The high count for misuse shows that AI risk comes from harmful use by people
as well as system error.
Affected stakeholder frequency
The general public was the most affected stakeholder group, with 60 cases. Customers or clients followed with
14 cases. Children or minors had 10 cases. Organizations and patients had 4 cases each. Vulnerable groups had
3 cases. Students or learners and individual users had 2 cases each, while employees or job applicants had 1 case.
This pattern shows that AI ethical risks often extend beyond direct users. Public-facing systems, synthetic media,
surveillance tools, and online platforms affect communities and public trust. The presence of minors in the
dataset also shows the need for stronger safeguards.
Ethical principle and severity frequency
Security was the most frequently violated ethical principle, with 28 cases. Transparency followed with 22 cases.
Safety and privacy had 12 cases each. Accountability and social responsibility had 11 cases each. Fairness had
4 cases.
Most incidents were rated High severity, with 75 cases. Critical incidents accounted for 21 cases, while Moderate
incidents accounted for 4 cases. No incident was rated Low severity. This pattern suggests that public AI incident
records often capture visible, serious, or widely reported harm.
Table 2 shows the cross-tabulation between ethical risk type and severity level.
Table 2. Ethical Risk Type by Severity Level
Ethical Risk Type
Moderate
High
Critical
Total
MISINFO - Misinformation or deception
4
29
0
33
Page 1112
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
MISUSE - Misuse or malicious use
0
24
2
26
SAFETY - Safety failure
0
6
6
12
PRIV - Privacy violation
0
3
9
12
ACC - Accountability failure
0
10
1
11
BIAS - Bias and discrimination
0
2
2
4
SEC - Security risk
0
1
1
2
Cross-tabulation patterns
Privacy violation had a high share of Critical severity cases. Safety failure also showed serious severity because
half of its cases were rated Critical. Bias and discrimination had fewer cases, but half were Critical. These
patterns show that privacy, safety, and fairness risks need close review even when they appear less often.
Generative AI and chatbots were often linked to misinformation or deception, accountability failure, and privacy
violation. Finance was strongly linked to misuse or malicious use.
Transportation was linked to safety failure. Education was linked mainly to privacy violation and safety failure.
Table 3 summarizes the dominant risk pattern by system domain.
Table 3. System Domain and Dominant Ethical Risk Pattern
System Domain
Frequency
Percent
Dominant Ethical Risk Pattern
GEN - Generative AI and
chatbots
28
28.0
MISINFO - Misinformation or deception (11); ACC -
Accountability failure (6); PRIV - Privacy violation (4)
LAW - Law enforcement
and surveillance
16
16.0
MISINFO - Misinformation or deception (5); MISUSE -
Misuse or malicious use (4); BIAS - Bias and
discrimination (4)
FIN - Finance
13
13.0
MISUSE - Misuse or malicious use (12); MISINFO -
Misinformation or deception (1)
GOV - Government and
public service
13
13.0
MISINFO - Misinformation or deception (7); MISUSE -
Misuse or malicious use (5); SAFETY - Safety failure (1)
EDU - Education
10
10.0
PRIV - Privacy violation (6); SAFETY - Safety failure (3);
MISINFO - Misinformation or deception (1)
SOC - Social media and
online platforms
8
8.0
MISINFO - Misinformation or deception (5); PRIV -
Privacy violation (2); MISUSE - Misuse or malicious use
(1)
TRN - Transportation
5
5.0
SAFETY - Safety failure (5)
Page 1113
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
BUS - Business and
customer service
4
4.0
MISINFO - Misinformation or deception (3); ACC -
Accountability failure (1)
HLT - Health care
3
3.0
ACC - Accountability failure (2); MISUSE - Misuse or
malicious use (1)
Deepfake-related ethical risk pattern
Deepfake-related incidents were included only as a supporting risk pattern under generative AI and synthetic
media systems. In the encoded dataset, 37 out of 100 records mentioned deepfake or synthetic media. These
cases were linked to misinformation, impersonation, fraud, privacy violation, reputational harm, and public
manipulation.
Deepfakes were not treated as the main focus of the study. They were included because several records showed
AI-generated or manipulated media being used to imitate real persons, distort information, support scams, or
target individuals. Prior studies on deepfakes report similar concerns involving deception, impersonation,
privacy harm, disinformation, and detection challenges [14, 17, 21, 27]. This keeps the study focused on AI-
based information systems while recognizing deepfakes as a recurring ethical risk.
DISCUSSION
The findings show that ethical risks in AI-based information systems appear across several domains. The results
support incident-based AI ethics research, which treats documented failures as useful evidence for risk analysis
[5, 9, 16, 23]. They also support sociotechnical harm frameworks, which explain AI risk through the interaction
of data, models, users, institutions, and affected communities [25, 26].
The high frequency of misinformation, deception, and misuse reflects the growth of generative AI and synthetic
media. These systems produce text, images, audio, video, and recommendations. Users sometimes treat their
outputs as factual. Institutions sometimes rely on automated content. Malicious actors use the same tools to
mislead the public, impersonate people, or commit fraud.
The high number of incidents affecting the general public shows that AI ethical risk is not limited to direct users.
Public-facing AI systems often affect communities, institutions, public trust, and vulnerable groups. This
supports the need for early screening before AI systems reach broad use.
The high and critical severity ratings also show the need for stronger controls. Privacy violations, safety failures,
child-related harms, fraud, surveillance, and public misinformation require careful review. These findings align
with risk-based governance approaches such as the NIST AI RMF and the EU AI Act [8, 20].
The findings also show that general AI ethics principles are not enough by themselves. Organizations need tools
that turn principles into operational questions. The screening matrix developed in this study addresses this need
by linking system domains, risk types, stakeholders, warning indicators, screening questions, and management
actions.
Generative AI and large language model studies support this result. These studies identify risks related to scale,
hallucination, misuse, user dependence, and downstream adaptation [4, 13, 19, 29]. The matrix helps
organizations examine these risks before deployment or expanded use.
Incident-Based Ethical Risk Screening Matrix
The Incident-Based Ethical Risk Screening Matrix was developed from the analysis of 100 publicly reported AI
incident records. It turns observed risk patterns into screening questions and management actions. The matrix
supports early ethical risk identification before AI systems are adopted, deployed, or expanded.
Page 1114
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
Table 4. Incident-Based Ethical Risk Screening Matrix for AI-Based Information Systems, Part 1: Risk
Identification
AI System Domain
Common Ethical Risks
Affected Stakeholders
Risk Level
Generative AI and
Chatbots
Misinformation, deception, privacy
violation, accountability failure, misuse
General public, students, users,
organizations
High
Deepfake and
Synthetic Media
Systems
Identity misuse, deception, fraud,
privacy violation, reputational harm
Individuals, public figures,
minors, organizations, general
public
High to
Critical
Law Enforcement and
Surveillance
Bias, discrimination, privacy violation,
lack of transparency, accountability
failure
General public, vulnerable
groups, government agencies,
individuals
Critical
Finance
Fraud, misuse, privacy violation,
security risk, accountability failure
Customers, clients, financial
institutions, general public
High
Education
Privacy violation, safety failure,
misinformation, overreliance, lack of
transparency
Students, teachers, schools,
children or minors
High
Health Care
Safety failure, privacy violation,
accountability failure, bias, lack of
transparency
Patients, health professionals,
health institutions
Critical
Government and
Public Service
Misinformation, privacy violation,
accountability failure, lack of
transparency, social harm
Citizens, public agencies,
vulnerable groups, general
public
High
Social Media and
Online Platforms
Misinformation, deception, misuse,
social harm, privacy violation
General public, minors,
platform users, vulnerable
groups
High
Employment and
Hiring
Bias, discrimination, lack of
transparency, privacy violation,
accountability failure
Job applicants, employees,
employers, vulnerable groups
High
Transportation
Safety failure, accountability failure,
security risk, lack of transparency
Passengers, pedestrians,
drivers, transport operators,
general public
Critical
Business and
Customer Service
Privacy violation, misinformation,
accountability failure, overreliance,
misuse
Customers, clients,
organizations, employees
Moderate to
High
Page 1115
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
Table 5. Incident-Based Ethical Risk Screening Matrix for AI-Based Information Systems, Part 2:
Screening and Management Actions
AI System
Domain
Warning Indicators
Screening Questions
Suggested Management
Action
Generative AI
and Chatbots
Generates text, images,
audio, or video that users
may treat as factual
information.
Is AI-generated content clearly
labeled? Is there human review
for sensitive outputs? Who is
accountable for harmful output?
Require human review.
Label AI content. Create
reporting channels. Limit
use in high-risk decisions.
Deepfake and
Synthetic Media
Systems
Imitates a real person’s
face, voice, image, or
likeness.
Was consent obtained? Is the
content labeled? Can the source
be verified? Is there a takedown
process?
Require consent
verification, source
verification, labels or
watermarks, rapid
takedown, and identity
protection.
Law
Enforcement and
Surveillance
Identifies, tracks, profiles,
scores, or monitors people.
Was the system tested for bias?
Is there human review? Is there
an appeal or correction process?
Require bias audit, strict
human oversight,
transparency reports, and
appeal mechanisms.
Finance
Handles identity,
transactions, credit
decisions, fraud detection,
or financial data.
Does the system protect
financial data? Is fraud
monitoring active? Can users
challenge automated decisions?
Strengthen identity
verification. Maintain
audit logs. Monitor fraud
patterns. Require human
review.
Education
Collects student data,
monitors learners, grades
outputs, or generates
academic content.
Does the system protect student
data? Are students informed of
AI use? Is there teacher review?
Apply data privacy
safeguards. Require
teacher oversight. Avoid
fully automated high-
stakes grading.
Health Care
Supports diagnosis, triage,
treatment recommendation,
patient monitoring, or
health records.
Does the system influence
medical decisions? Was it
clinically validated? Who is
responsible for wrong
recommendations?
Require medical expert
review, clinical validation,
patient data protection, and
audit trails.
Government and
Public Service
Affects public services,
identity verification,
eligibility, benefits, or
public information.
Can citizens understand and
challenge decisions? Is public
data protected? Is there a
correction process?
Require transparency
review, appeal
mechanisms, human
validation, and public data
safeguards.
Social Media and
Online Platforms
Recommends content,
moderates posts, generates
synthetic media, or
influences public opinion.
Does the system amplify
harmful content? Are AI-
generated materials labeled?
Are minors protected?
Improve moderation.
Label synthetic content.
Add reporting channels.
Protect minors and
vulnerable users.
Employment and
Hiring
Screens applicants, ranks
resumes, monitors workers,
Was the system tested for bias?
Can applicants appeal? Are
Require bias audit. Keep
human decision authority.
Page 1116
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
or scores employee
performance.
decision criteria explainable?
Provide appeal options.
Limit unnecessary
surveillance.
Transportation
Supports autonomous
driving, routing, navigation,
traffic decisions, or safety-
critical movement.
Can system failure cause
physical harm? Is there human
override? Who is responsible
during failure?
Require safety testing,
emergency protocols,
human override, and clear
accountability.
Business and
Customer Service
Handles customer data,
service requests, product
recommendations, or
automated responses.
Are customers informed when
they interact with AI? Can
customers reach a human agent?
Is customer data protected?
Disclose AI use. Protect
customer data. Review AI
responses. Provide human
support.
Together, Tables 4 and 5 present the screening matrix in portrait format. Table 4 shows risk identification. Table
5 shows screening questions and management actions.
CONCLUSION AND RECOMMENDATIONS
Conclusion
This study developed an Incident-Based Ethical Risk Screening Matrix for AI-Based Information Systems using
100 publicly reported AI incident records from 2020 to 2026. The incidents were classified by system domain,
ethical risk type, affected stakeholder, ethical principle violated, severity level, and evidence strength.
The findings show that ethical risks in AI-based information systems recur across different domains. The most
common risks were misinformation or deception, misuse or malicious use, safety failure, privacy violation, and
accountability failure. These risks appeared most often in generative AI and chatbots, law enforcement and
surveillance, finance, government and public service, education, and social media platforms.
The general public was the most affected stakeholder group. This means that AI risk often reaches beyond direct
users and affects communities, institutions, public trust, and vulnerable groups. Incidents involving minors,
health care, transportation, surveillance, and public services raised stronger severity concerns because they
involve safety, privacy, rights, and social accountability.
Most cases were rated High severity. A smaller but important group was rated Critical. This pattern suggests
that public AI incident reporting often captures cases with visible harm, public concern, or broad impact.
The study concludes that AI-based information systems need structured ethical screening before adoption,
deployment, or expansion. The proposed matrix offers a practical guide for identifying warning indicators,
asking screening questions, and selecting management actions.
Recommendations
1. Organizations should conduct ethical risk screening before deploying AI-based information systems. They
should check fairness, privacy, transparency, accountability, safety, security, human oversight, and social
responsibility.
2. IT managers should use the Ethical Risk Screening Matrix during system planning and procurement. The
matrix supports early risk identification before AI systems are acquired, developed, or integrated.
3. High-risk AI domains should undergo stricter review. AI systems used in health care, law enforcement,
surveillance, transportation, finance, education, and public service should receive stronger ethical assessment.
Page 1117
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
4. Organizations should require human oversight for high-impact AI decisions. AI systems that affect
employment, education, health, financial access, public benefits, safety, or legal outcomes should not operate
without human review.
5. AI-generated content should be labeled and reviewed in sensitive contexts. Generative AI systems should
include disclosure mechanisms, content review, and error reporting channels.
6. Deepfake and synthetic media risks should be controlled through consent verification, source verification,
labeling, identity protection, and rapid takedown procedures.
7. Organizations should protect vulnerable groups and minors. Privacy protection, consent, safety checks, and
human intervention should receive priority.
8. AI incident reporting should be strengthened. Organizations should document AI-related errors, complaints,
harms, and near-harms to support learning and prevention.
9. Future researchers should expand the dataset. Future studies should compare AI incident databases or focus
on domains such as education, health care, finance, or generative AI.
10. Future studies should validate the screening matrix through expert review or organizational case studies.
Overall, the study recommends treating AI ethics as a management responsibility. AI systems should be assessed
based on their effect on people, organizations, rights, safety, privacy, and public trust.
Data availability
The data used in this study came from publicly available AI incident records. The coded dataset used for the
analysis is available in the researchers' working file and will be shared for academic review if required.
Conflict of interest
The authors declare no conflict of interest.
REFERENCES
1. AI Incident Database. (n.d.). Welcome to the Artificial Intelligence Incident Database.
https://incidentdatabase.ai/
2. Atherton, D. (2026, February 2). AI Incident Roundup: November and December 2025 and January 2026.
AI Incident Database. https://incidentdatabase.ai/blog/incident-report-2025-november-december-2026-
january/
3. Autio, C., Schwartz, R., Dunietz, J., Jain, S., Stanley, M., Tabassi, E., Hall, P., & Roberts, K. (2024).
Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI
600-1). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.AI.600-1
4. Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic
parrots: Can language models be too big? Proceedings of the 2021 ACM Conference on Fairness,
Accountability, and Transparency, 610-623. https://doi.org/10.1145/3442188.3445922
5. Burema, D., Debowski-Weimann, N., von Janowski, A., Grabowski, J., Maftei, M., Jacobs, M., van der
Smagt, P., & Benbouzid, D. (2023). A sector-based approach to AI ethics: Understanding ethical issues of
AI-related incidents within their sectoral context. Proceedings of the 2023 AAAI/ACM Conference on AI,
Ethics, and Society, 705–714. https://doi.org/10.1145/3600211.3604680
6. Diel, A., Lalgi, T., Mellis, F. S., Teufel, M., & Bäuerle, A. (2025). The harm of deepfakes: A scoping
review of deepfakes’ negative effects on human mind and behavior. AI & Society.
https://doi.org/10.1007/s00146-025-02774-0
7. Dixon, R. B. L., & Frase, H. (2025). AI incidents: Key components for a mandatory reporting regime.
Page 1118
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
Center for Security and Emerging Technology. https://doi.org/10.51593/20240023
8. European Parliament & Council of the European Union. (2024). Regulation (EU) 2024/1689 laying down
harmonised rules on artificial intelligence. Official Journal of the European Union. https://eur-
lex.europa.eu/eli/reg/2024/1689/oj
9. Feffer, M., Martelaro, N., & Heidari, H. (2023). The AI Incident Database as an educational tool to raise
awareness of AI harms: A classroom exploration of efficacy, limitations, and future improvements.
Proceedings of the 3rd ACM Conference on Equity and Access in Algorithms, Mechanisms, and
Optimization, Article 3. https://doi.org/10.1145/3617694.3623223
10. Gabriel, I. (2020). Artificial intelligence, values, and alignment. Minds and Machines, 30, 411-437.
https://doi.org/10.1007/s11023-020-09539-2
11. Hadan, H., Mogavi, R. H., Zhang-Kennedy, L., & Nacke, L. E. (2025). Who is responsible when AI fails?
Mapping causes, entities, and consequences of AI privacy and ethical incidents. International Journal of
Human-Computer Interaction. https://doi.org/10.1080/10447318.2025.2549073
12. Helmus, T. C. (2022). Artificial intelligence, deepfakes, and disinformation: A primer. RAND
Corporation. https://doi.org/10.7249/PEA1043-1
13. Ji, Z., Lee, N., Frieske, R. M., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023).
Survey of hallucination in natural language generation. ACM Computing Surveys, 55(12), Article 248.
https://doi.org/10.1145/3571730
14. Kietzmann, J., Lee, L. W., McCarthy, I. P., & Kietzmann, T. C. (2020). Deepfakes: Trick or treat? Business
Horizons, 63(2), 135-146. https://doi.org/10.1016/j.bushor.2019.11.006
15. Knight, S., McGrath, C., Viberg, O., & Cerratto Pargman, T. (2025). Learning about AI ethics from cases:
A scoping review of AI incident repositories and cases. AI and Ethics, 5, 2037-2053.
https://doi.org/10.1007/s43681-024-00639-8
16. McGregor, S. (2021). Preventing repeated real-world AI failures by cataloging incidents: The AI Incident
Database. Proceedings of the AAAI Conference on Artificial Intelligence, 35(17), 15458-15463.
https://doi.org/10.1609/aaai.v35i17.17817
17. Mirsky, Y., & Lee, W. (2021). The creation and detection of deepfakes: A survey. ACM Computing
Surveys, 54(1), Article 7. https://doi.org/10.1145/3425780
18. Mökander, J., & Floridi, L. (2021). Ethics-based auditing to develop trustworthy AI. Minds and Machines,
31(2), 323–327. https://doi.org/10.1007/s11023-021-09557-8
19. Mökander, J., Schuett, J., Kirk, H. R., & Floridi, L. (2024). Auditing large language models: A three-
layered approach. AI and Ethics, 4, 1085-1115. https://doi.org/10.1007/s43681-023-00289-2
20. National Institute of Standards and Technology. (2023). Artificial Intelligence Risk Management
Framework (AI RMF 1.0) (NIST AI 100-1). U.S. Department of Commerce.
https://doi.org/10.6028/NIST.AI.100-1
21. Nguyen, T. T., Nguyen, Q. V. H., Nguyen, D. T., Nguyen, D. T., Huynh-The, T., Nahavandi, S., Nguyen,
T. T., Pham, Q. V., & Nguyen, C. M. (2022). Deep learning for deepfakes creation and detection: A survey.
Computer Vision and Image Understanding, 223, Article 103525.
https://doi.org/10.1016/j.cviu.2022.103525
22. Organisation for Economic Co-operation and Development. (2025). Towards a common reporting
framework for AI incidents (OECD Artificial Intelligence Papers, No. 34). OECD Publishing.
https://doi.org/10.1787/f326d4ac-en
23. Paeth, K., Atherton, D., Pittaras, N., Frase, H., & McGregor, S. (2025). Lessons for editors of AI incidents
from the AI Incident Database. Proceedings of the AAAI Conference on Artificial Intelligence, 39(28),
28946-28953. https://doi.org/10.1609/aaai.v39i28.35163
24. Raji, I. D., Smart, A., White, R. N., Mitchell, M., Gebru, T., Hutchinson, B., Smith-Loud, J., Theron, D.,
& Barnes, P. (2020). Closing the AI accountability gap: Defining an end-to-end framework for internal
algorithmic auditing. In Proceedings of the 2020 Conference on Fairness, Accountability, and
Transparency (FAT* ’20) (pp. 33–44). Association for Computing Machinery.
https://doi.org/10.1145/3351095.3372873
25. Shelby, R., Rismani, S., Henne, K., Moon, A., Rostamzadeh, N., Nicholas, P., Yilla-Akbari, N., Gallegos,
J., Smart, A., Garcia, E., & Virk, G. (2023). Sociotechnical harms of algorithmic systems: Scoping a
taxonomy for harm reduction. Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and
Page 1119
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue VI, June 2026
Society, 723-741. https://doi.org/10.1145/3600211.3604673
26. Slattery, P., Saeri, A. K., Grundy, E. A. C., Graham, J., Noetel, M., Uuk, R., Dao, J., Pour, S., Casper, S.,
& Thompson, N. (2024). The AI Risk Repository: A comprehensive meta-review, database, and taxonomy
of risks from artificial intelligence. arXiv. https://doi.org/10.48550/arXiv.2408.12622
27. Tolosana, R., Vera-Rodriguez, R., Fierrez, J., Morales, A., & Ortega-Garcia, J. (2020). Deepfakes and
beyond: A survey of face manipulation and fake detection. Information Fusion, 64, 131-148.
https://doi.org/10.1016/j.inffus.2020.06.014
28. Vaccari, C., & Chadwick, A. (2020). Deepfakes and disinformation: Exploring the impact of synthetic
political video on deception, uncertainty, and trust in news. Social Media + Society, 6(1).
https://doi.org/10.1177/2056305120903408
29. Weidinger, L., Uesato, J., Rauh, M., Griffin, C., Huang, P.-S., Mellor, J., Glaese, A., Cheng, M., Balle, B.,
Kasirzadeh, A., Biles, C., Brown, S., Kenton, Z., Hawkins, W., Stepleton, T., Birhane, A., Hendricks, L.
A., Rimell, L., Isaac, W., ... Gabriel, I. (2022). Taxonomy of risks posed by language models. Proceedings
of the 2022 ACM Conference on Fairness, Accountability, and Transparency, 214-229.
https://doi.org/10.1145/3531146.3533088
30. Wei, M., & Zhou, Z. (2022). AI ethics issues in real world: Evidence from AI Incident Database. arXiv.
https://doi.org/10.48550/arXiv.2206.07635
